A Fast UAV Aerial Image Mosaic Method Based on Improved KAZE

This paper presents a novel image mosaic method for the unmanned aerial vehicle (UAV) aerial image based on the improved KAZE algorithm. For practical implementation and cost efficiency, the matching speed of KAZE algorithm is not desirable and random sample consensus (RANSAC) algorithm takes plenty of time to screen the correct feature pairs. In addition, the UAV aerial image is susceptible to environmental and other factors, such as scale, noise and brightness change. To overcome these problems, the novel image mosaic method is proposed. In particular, the non-linear scale space is constructed using the fast-explicit diffusion (FED) algorithm. Then, the feature points are described by the improved fast retina key-point (FREAK) feature descriptor. Further, we adopt the Hamming algorithm for a rough match of these feature points, and select the improved progressive sample consensus (PROSAC) algorithm to exact correct matches. Finally, the image is stitched by the weighted average algorithm. Results from numerical experiments show that compared with the scale invariant feature transform (SIFT), speed-up robust features (SURF), oriented FAST and rotated BRIEF (ORB) as well as the KAZE algorithm, the proposed algorithm has better performance on feature extraction speed, feature matching speed, as well as matching accuracy.